Skip to content

A macOS app to parse face landmarks from a video for training GANs

License

Notifications You must be signed in to change notification settings

xiaohk/FaceData

Repository files navigation

Face Data

A macOS application used to auto-annotate landmarks from a video. Those landmarks can further be used as training data for Generative Adversarial Networks (GANs).

DOI License

Getting Started

Installing

You can either download the binary file from Rease or build the source code using Xcode.

Use

Description
Video Path Path to the video file, currently only support .mp4 files. Use Select File to generate path using a file browsing panel.
Output Path Path to the output directory, this app will create origin and landmarks two sub-directories. Use Select Folder to generate path using a file browsing panel.
Start Second An integer value indicating from which second to start capturing frames from the video, default is 0 (from the beginning)
End Second This app would not extract frames after this second. Default is the duration of the video.
# of Frames Integer value of how many frames you want to generate. Default is 100 frames.
Start Start the process.
Cancel Stop the process.

Output

  • Two sub-directories origin and landmark will be created in the specified output directory.
  • origin contains the original frames extracted from the video, with file name: img001.png.
  • landmark contains the landmark image drawn based on the corresponding frame in origin, with file name: img001lm.png.
  • If there is no face detected in one original frame, the corresponding file name in landmark is no_face_img001lm.png.

Output Images Processing

You will probably want to process the generated images to fit the size restriction for you GANs model. You can refer the Python script crop.py.

Built With